Yesterday, AMD confirmed what had been an open secret that the Radeon RX 8000 would be commitment to mid-range graphics cards, and trying to offer as close to the price/performance ratio as possible. The company has adopted this strategy because it believes it is the one that can work best to meet a clear objective and increase its market share to 4-50%.

It’s the goal very ambitious and also quite complicatedespecially considering the dominance NVIDIA has right now and the competition Intel could offer with Arc Battlemage, but AMD is determined and willing to fight to achieve it, so we’ll have to wait a fair amount of time to see what results it gets. with this new strategy.

There’s also been new information about the Radeon RX 8000 coming out over the last few weeks, so I think it’s a good time to post new article updated with everything we know right now about this new generation of graphics cards. All the information that I will share with you comes from fairly reliable sources, so it has high credibility, but since it is not official, it is not completely guaranteed that it will be fulfilled.

RDNA 4 architecture details

The RDNA 4 architecture will be the foundation on which the Radeon RX 8000 GPUs will be built, and if all the information I’ve seen comes to pass, it will mean a very important development versus RDNA 3.

The first thing to note is that this new architecture could mean abandoning the MCM design we saw in the Radeon RX 6000 and RX 7000, meaning that AMD will go back to monolithic core design with the Radeon RX 8000, where there would be no externalized components and all the GPU elements would be on the same silicon chip, including the L3 cache.

On a Radeon RX 6700 increasing latency in the communication between the GPU and the specified cache. Integrating the L3 cache into the same silicon as the GPU eliminates this problem and improves performance.

In order to integrate the L3 cache on the same silicon chip, AMD will use it TSMC 4nm node, which reduces the size of the transistors and improves the efficiency and will reduce its capacity below 96 MB. It is very likely that the most powerful models will be 64 MB and the least powerful between 48 MB and 32 MB.

This architecture could double the number of cores to speed up ray tracing per computing unit, which would mean going from one to two, and will bring new technologies and functions to improve performance in ray tracing, among which we can highlight the engine for the intersection of two rays, improvements to work with hierarchies of boundary volumes and three-pair ray tracing optimization.

RDNA 4 will also offer performance improvements in rasterization and AI due to rollout field acceleration hardware and FP8 operations. It’s not clear yet, but this hardware could be the basis for AMD to develop a new intelligent upscaling technology that is truly on par with NVIDIA DLSS.

AMD will launch four models under Radeon RX 8000

And these four models will use two graphics cores with different configurations. Two models will use the core Navi 48which is the most powerful, and the other two will be based on the graphics core Navi 44. To shape different graphics cards, AMD adjusts the active computing units in each of these graphics cores.

Radeon RX 8800 XT will be the most powerful of these four new graphics cards. Behind him will be the Radeon RX 8700 XT and in the economic middle class we will find the Radeon RX 8600 XT and the Radeon RX 8500. AMD may use a different nomenclature and the two most powerful models can be identified as the Radeon RX 8700 XT and the Radeon RX 8700.

Possible specifications of Radeon RX 8800 XT

• Navi 48 graphics core made on TSMC’s 4nm node.
• 64 computing units.
• 4,096 shaders at over 3 GHz.
• 256 texturing units.
• 128 raster units.
• 128 third-generation cores to accelerate ray tracing.
• 128 second generation cores for AI.
• 256-bit bus.
• 16 GB of GDDR6 memory at 20 Gbps (640 GB/s bandwidth).
• 4MB L2 cache.
• 64 MB L3 cache.

Comparative strength and performance

A Radeon RX 8800 XT would have the performance 49.15 TFLOPs in FP32which would put it slightly above the 48.74 TFLOPs it achieved GeForce RTX 4080. In ray tracing, its performance could be similar to that of the GeForce RTX 4070 Ti.

Possible specifications of Radeon RX 8700 XT

• Navi 48 graphics core made on TSMC’s 4nm node.
• 56 computing units.
• 3,584 shaders at 3 GHz.
• 224 texturing units.
• 128 raster units.
• 112 third-generation cores to accelerate ray tracing.
• 112 second generation cores for AI.
• 256-bit bus.
• 16 GB of GDDR6 memory at 20 Gbps (640 GB/s bandwidth).
• 4MB L2 cache.
• 64 MB L3 cache.

Comparative strength and performance

This model will have power 43 TFLOPs in FP32enough to get me high between GeForce RTX 4070 SUPER and GeForce RTX 4070 Ti. Its ray tracing performance could be similar to that of the GeForce RTX 4070.

Possible specifications of Radeon RX 8600 XT

• Navi 44 graphics core made on TSMC’s 4nm node.
• 40 computing units.
• 2,560 shaders at speeds up to 3.2 GHz.
• 160 texturing units.
• 64 raster units.
• 80 third-generation cores to accelerate ray tracing.
• 80 second generation cores for AI.
• 192-bit bus.
• 12 GB of GDDR6 memory at 18 Gbps (432 GB/s bandwidth).
• 2 MB L2 cache.
• 48 MB L3 cache.

Comparative strength and performance

It could be the big star of AMD’s new generation in price-performance ratio. His strength in FP32 will be 30.72 TFLOPs, the character that places it on the level GeForce RTX 4070. Its ray tracing performance should be closer to that of the GeForce RTX 4060 Ti.

Possible specifications of Radeon RX 8500 XT

• Navi 44 graphics core made on TSMC’s 4nm node.
• 32 computing units.
• 2048 shaders at speeds up to 3 GHz.
• 128 texturing units.
• 64 raster units.
• 64 third-generation cores to accelerate ray tracing.
• 64 second generation cores for AI.
• 128-bit bus.
• 8 GB of GDDR6 memory at 18 Gbps (288 GB/s bandwidth).
• 2 MB L2 cache.
• 32 MB of L3 cache.

Comparative strength and performance

It will be the most economical of these four new graphics cards. It will have power 24.57 TFLOPs, a character that would leave her a little above GeForce RTX 4060 Ti. In ray tracing, it should be more or less at the level of the GeForce RTX 4060.

Keep in mind that TFLOPs allow us to make rough estimates of performance, but that they are not absolute values ​​when talking about games, and therefore performance in these types of applications may vary, because it depends on many other factors, including things like resolution, bandwidth, and optimization.

Possible release date and price

No deal has yet been reached, but the latest information says that the Navi 48 will enter the development phase series production during the fourth quarter this year, so we have two options:

• That the Radeon RX 8800 XT and RX 8700 XT will arrive before the end of 2024, with availability possibly a bit limited
• Let both graphics cards retire in early 2025.

In both cases, the Radeon RX 8600 XT and Radeon RX 8500 XT seem to arrive later than the more powerful models. If true, they won’t be available until first or second quarter of 2025.

Regarding the price, we know that AMD wants to fine-tune the price-performance ratio with the Radeon RX 8000 and it is rumored that the most powerful model It will cost between $499 and $599. This allows me to make an estimate possible prices of all models of this new generation:

• Radeon RX 8800 XT: between $499 and $599. Less than 649 euros depending on the exchange rate.
• Radeon RX 8700 XT: between $399 and $499. Less than 549 euros depending on the exchange rate.
• Radeon RX 8600 XT: between $299 and $349. In exchange for less than 400 euros.
• Radeon RX 8500 XT: between $199 and $249. In exchange for less than 300 euros.

AI generated images.